The Voynich Manuscript is undeciphered. This project measures competing explanations with the same statistics as the manuscript: verbose ciphers in several plaintext languages, a Hebrew reading, meaningless generators. Every reading is tested against control texts that carry no message.
A handwritten parchment codex of around 240 surviving pages, originally an estimated 272 pages in 20 quires. It contains botanical sections with over 100 unidentified plant drawings, astronomical-astrological diagrams with zodiac signs, balneological illustrations of bathing figures in tube systems, cosmological fold-out pages, pharmaceutical sections, and a continuous text section without illustrations. The entire text is written in an undeciphered script of about 20 to 25 recurring signs, known as “Voynichese”.
From 15th-century parchment workshops to digital analysis in the 21st century.
Radiocarbon dating of four parchment samples (folios 8, 26, 47, 68) at the University of Arizona (2009) consistently yields this period. Ink and pigment analyses fit the early 15th century. A modern forgery is therefore unlikely. Stylistic features of the illustrations point to a place of origin in northern Italy.
A surviving accompanying letter names Emperor Rudolf II as an early owner, supposedly acquired for 600 ducats. Its earlier provenance remains unknown; the ownership record between the 17th and early 20th century is poorly documented.
The antiquarian Wilfrid M. Voynich acquires the codex from the holdings of the Jesuit College of Villa Mondragone near Frascati. A sale at the demanded price of around 100,000 US dollars is not concluded during his lifetime.
After Voynich’s death the manuscript stays first with Ethel Voynich, later in the estate of her secretary, before being passed on through the antiquarian H. P. Kraus.
Transfer to the Beinecke Rare Book & Manuscript Library at Yale University. The digitized pages have since counted among the library’s most-viewed holdings. The text remains undeciphered.
The rules every model is tested by, whichever explanation it stands for.
No explanation is taken as given. Cipher, language and random models are measured with the same statistics as the manuscript: entropy, positional binding of the glyphs, Zipf curve, share of one-off words, word length. The Hebrew reading model behind the lexicon, grammar and translator is one of them.
A reading only counts as evidence once it reads the manuscript better than texts without a message: Markov text built from Voynich glyphs, ciphertext of a Latin plaintext, words with shuffled glyphs. What the controls “read” just as well says nothing about the meaning of the manuscript.
For pre-registered tests, design, texts and decision rule are fixed before measuring and frozen by a hash. Results are published even when they count against a model, the Hebrew one included.
Botanical analyses follow a fixed order: first independent image observation along all morphological features, then one’s own botanical candidates. The Petersen reference list is consulted only at the end, for comparison, never as a starting hypothesis.
The best-verified sequences in the corpus, as a guide for translating.
Identical l-directional morphology, opposite semantics. Major new finding of v5.2. lor is widely attested (27 folios); lshol so far attested only as a single occurrence (f7v). ★★★★
First attestation of ykchor (he heals the pupil; 4th verb in the y+k+[terminus] paradigm) and of ycheor= as a positive closing colophon. f11v (Borago type) is the most positive prognostic folio in Quire B. odl (9 attestations/4 folios) is not yet formalized as its own lexicon entry. schold at f11v P.6 is a 2/4 segmentation tie (H/F: schold; C/U: s + chold). Both readings are equally well attested. ★★★
ykaiin completes the y+k+[terminus] paradigm to five clinical verb forms (life, light, pupil, pallor, eye). The blood triad dol·dair·dam (f14v P.8) is the first attestation of dam in Quire B and connects f14v with f3r (blood plant). ★★★★★
soloiin (f15v P.11) = “medicine for the eye”, the most therapeutically precise compound in the corpus. ykshol (f15r P.12) = first negative y+k+terminus; the paradigm now has 6 members plus candidate ykair. Bifolio bB2 (f10r/f10v/f15r/f15v) closes a consistent ophthalmological program. qotchor is subject to an ongoing OP-A review of its prefix classification; the translation stands until that decision. ★★★★★
tchor (תְּ+חֹר): fem. pupil-healing verb, two attestations confirmed in bB1 ★★★★. chkar·chkar (f16v P.6): first double examination formula in the corpus. Bifolio bB1 (f9r/f9v/f16r/f16v) completes Quire B, the first purely positive quire ending in the analyzed corpus. sokar and oaorar are so far attested only on f16v (candidate status ★★, further attestation pending). ★★★
f18r: Centaurea montana, strongest closing colophon in Quire C; dar·dar (R36 first attestation), doldaiin (judgment of illness, corpus first attestation). f18v: Dracunculus vulgaris, first zoomorphic companion figure in the botanical corpus; ykam/ytor resurrection verbs; qokchy·qokchy R37 first attestation. R39 (dom=) still carries candidate status (R43 deadline open). yko (5 attestations/2 folios) is not yet formalized as its own lexicon entry. ★★★
Three independently confirmed words + Aramaic genitive suffix in natural word order. sar and al are majority-consented here; at f57r P.9 daiindy is read that way by only one of three transcriber layers (F) (H/C: uncertain character at the same position). At f100r (4/4) and f105r (3/3) the word itself has majority consensus. ★★★★
First complete zodiac register: R17 ★★★★ (5th register), R19/R35 ★★★ (zodiac first attestation), R50 ★★★★ (2nd register), R52a ★★★★ (4 folios · 75 labels · 47%). alef/otalef = first iconographic letter reference in the zodiac corpus. ★★★★
Line-end formula: binary prognosis (healing or no way out), identical to the Hippocratic-Arabic prognosticon schema.
Complete Hippocratic prognosis schema in five words. sheckhy (124 attestations/19 folios) is broadly attested but not yet in the lexicon, making it a priority candidate for formalization. ★★★★
Triple prognosis: exhaustion → death → patient dying. chalal appears exclusively after sheol.
Identical l-directional morphology, opposite semantics. Major new finding of v5.2. lor is widely attested (27 folios); lshol so far attested only as a single occurrence (f7v). ★★★★
First attestation of ykchor (he heals the pupil; 4th verb in the y+k+[terminus] paradigm) and of ycheor= as a positive closing colophon. f11v (Borago type) is the most positive prognostic folio in Quire B. odl (9 attestations/4 folios) is not yet formalized as its own lexicon entry. schold at f11v P.6 is a 2/4 segmentation tie (H/F: schold; C/U: s + chold). Both readings are equally well attested. ★★★
ykaiin completes the y+k+[terminus] paradigm to five clinical verb forms (life, light, pupil, pallor, eye). The blood triad dol·dair·dam (f14v P.8) is the first attestation of dam in Quire B and connects f14v with f3r (blood plant). ★★★★★
soloiin (f15v P.11) = “medicine for the eye”, the most therapeutically precise compound in the corpus. ykshol (f15r P.12) = first negative y+k+terminus; the paradigm now has 6 members plus candidate ykair. Bifolio bB2 (f10r/f10v/f15r/f15v) closes a consistent ophthalmological program. qotchor is subject to an ongoing OP-A review of its prefix classification; the translation stands until that decision. ★★★★★
tchor (תְּ+חֹר): fem. pupil-healing verb, two attestations confirmed in bB1 ★★★★. chkar·chkar (f16v P.6): first double examination formula in the corpus. Bifolio bB1 (f9r/f9v/f16r/f16v) completes Quire B, the first purely positive quire ending in the analyzed corpus. sokar and oaorar are so far attested only on f16v (candidate status ★★, further attestation pending). ★★★
f18r: Centaurea montana, strongest closing colophon in Quire C; dar·dar (R36 first attestation), doldaiin (judgment of illness, corpus first attestation). f18v: Dracunculus vulgaris, first zoomorphic companion figure in the botanical corpus; ykam/ytor resurrection verbs; qokchy·qokchy R37 first attestation. R39 (dom=) still carries candidate status (R43 deadline open). yko (5 attestations/2 folios) is not yet formalized as its own lexicon entry. ★★★
Three independently confirmed words + Aramaic genitive suffix in natural word order. sar and al are majority-consented here; at f57r P.9 daiindy is read that way by only one of three transcriber layers (F) (H/C: uncertain character at the same position). At f100r (4/4) and f105r (3/3) the word itself has majority consensus. ★★★★
First complete zodiac register: R17 ★★★★ (5th register), R19/R35 ★★★ (zodiac first attestation), R50 ★★★★ (2nd register), R52a ★★★★ (4 folios · 75 labels · 47%). alef/otalef = first iconographic letter reference in the zodiac corpus. ★★★★
Line-end formula: binary prognosis (healing or no way out), identical to the Hippocratic-Arabic prognosticon schema.
Complete Hippocratic prognosis schema in five words. sheckhy (124 attestations/19 folios) is broadly attested but not yet in the lexicon, making it a priority candidate for formalization. ★★★★
Triple prognosis: exhaustion → death → patient dying. chalal appears exclusively after sheol.
Enter EVA text into the input field (words separated by spaces or ·; a new line = a new phrase, = marks a paragraph end). The tool searches the confirmed lexicon, recognizes prefixes, and shows Hebrew, a word-by-word analysis, and the German meaning. Unknown words are marked as such.
The tool works on the confirmed lexicon (419 entries, ★★★ or higher). Complex compounds and unknown words are marked as “unknown”. The output is a hypothesis, not a finished translation. The same pipeline “reads” meaningless Markov text and Latin ciphertext almost as often as the manuscript, see the model comparison.
None of the explanations tested is proven. So far the Hebrew readings do not tell the manuscript apart from meaningless control texts, neither at word nor at sentence level; the lexicon, grammar and translator therefore document work in progress, not a decipherment. A verbose cipher explains the manuscript’s low entropy and positional binding with every plaintext language tested. But no model yet explains the high share of one-off words together with the steep Zipf curve.
→ View all open problems